C-SURE: Shrinkage Estimator and Prototype Classifier for Complex-Valued Deep Learning

Rudrasis Chakraborty, Yifei Xing, Minxuan Duan, Stella X. Yu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 80-81

The James-Stein shrinkage estimator is a biased estimator that captures the mean of Gaussian random vectors. Recognized by its dominance over the maximum likelihood estimator (MLE) in terms of mean squared error (MSE), the James-Stein estimator has gained huge interests from the statistical field. However, little progress is made for extending the estimator onto complex manifold-valued data. In this work, we propose a novel Stein's unbiased risk estimator (SURE) on the complex field with theoretically proven optimum over the MLE. We empirically compare and analyze results of our proposed model on a publicly available complex-valued dataset where we can achieve better results than other state-of-the-art methods.